There are many factors that affect the decisions of respondents to cooperate when they are sampled for a survey. To the extent a researcher can accurately predict a given respondent's likelihood to cooperate, i.e., predict the respondent's individual "response propensity" (cf. Groves et al, 2004), then in theory, the researcher could target differential approaches to recruiting each respondent rather than merely utilizing a "one size fits all" approach. Our previous findings from a large mail survey (n=10,000) conducted in January 2004 (cf. Lavrakas et al, 2004) and a larger national mail survey (n=69,000) conducted in May 2004 (cf. Burks, et al, 2004) showed that cooperation in the mail survey could be predicted at a level significantly better than chance by using data about the (1) mailed respondent/household gathered in a prior telephone survey, (2) variables associated with the calling history of the respondent's telephone number, (3) interviewer ratings about the respondent, and (4) Census variables at the level of the local zipcode where the respondent lived. Our next step in this line of research is to use Census data at the block group level to determine what gains can be achieved in predicting response propensity to a mail survey among a national sample of 18-34 year olds (n=1,000). Our previous two studies showed an ability to predict cooperation in a mail survey among this age cohort with approximately 60% accuracy. Bringing the level of aggregation for the geographic variables in our model to a smaller level (block group) is more precise in nature, and is expected to contribute to improving the predictive power of our statistical model. The implications of these findings will be discussed as they relate to the differential deployment of finite resources for incenting households and strategies to increase response rates.